Prediction Of Laminar Burning Velocity Of C₁-C₄ Alkane Fuel Using Machine Learning

Accurate laminar combustion velocity prediction model can provide input parameters for engine numerical simulation,which is of great significance for the design and improvement of internal combustion engine.CHEMKIN numerical simulation software combined with detailed chemical reaction kinetics mechanism can get accurate laminar combustion velocity prediction results,but the cost of coupling calculation with engine simulation software is too high,and it is difficult to achieve ideal results by simplifying the mechanism.Based on a large number of numerical simulation data of C₁-C₄alkane fuel laminar combustion velocity,the prediction models of C₁-C₄alkane fuel laminar combustion velocity are established by using machine learning multivariable linear regression,neural network and support vector machine algorithm,and the prediction ability and adaptability of the models are compared and evaluated.The three models can quickly and accurately predict C₁-C₄alkane fuel laminar combustion velocity under wide conditions,and can provide simple and accurate laminar combustion velocity input parameters for engine numerical simulation.The specific research contents and main conclusions are as follows:1)The laminar combustion velocity of methane,ethane,propane and butane were simulated by CHEMKINⅡsoftware package and USC mech 2.0,and a multivariate linear regression sample database was established.Based on the empirical formula of laminar combustion velocity and multivariable linear regression algorithm,multivariable linear regression models for predicting laminar combustion velocity of methane,ethane,propane and butane were established.The fitting degree of the models is better than 0.99.2)Based on the established laminar combustion velocity model of C₁-C₄alkane fuel,the effects of initial temperature and pressure on the laminar combustion velocity of C₁-C₄alkane fuel are analyzed.The laminar combustion velocity of C₁-C₄alkane fuel has a positive exponential relationship with initial temperature and a negative exponential relationship with initial pressure.In the equivalence ratio range of 0.7-1.4,the temperature index of methane model is greater than that of ethane,propane and butane,which indicates that the laminar combustion velocity of methane is the most sensitive to temperature change,while the laminar combustion velocity of ethane,propane and butane has little difference to temperature change;the relationship of pressure index is CH₄>C₃H₈>C₄H₁₀>C₂H₆,which indicates that the laminar combustion velocity of methane is the most sensitive to the pressure change in C₁-C₄alkane fuel,and the laminar combustion velocity of ethane is the least sensitive to the pressure change.3)Combined with the fuel characteristics of C₁-C₄alkane fuel,the multivariable linear regression database was expanded,and the support vector machine and neural network databases were established.The laminar combustion velocity prediction model of C₁-C₄alkane fuel was established by using BP neural network and support vector machine algorithm respectively.The fitting degree of BP neural network model and support vector machine model is 0.9967 and 0.9952 respectively,which indicates that the two models have good prediction performance and can accurately predict the laminar combustion velocity of C₁-C₄alkane fuel.4)Through the research on the prediction ability of the three models,it is found that the three models can get better prediction results of the extraterritorial data,which indicates that the models have better generalization ability.